I haven't found any obvious source in aufs for the extra fputs which I suspect are causing this problem. If you could either give me more information so I can run the same tests myself (I'm guessing the problem isn't arch-specific) or else reproduce the problem with the kernel patched with something like the following, perhaps we can catch it in the act. However we also might just catch legitimate fputs since the erroneous ones could occur while the refcount is still positive ...
diff --git a/fs/file_table.c b/fs/file_table.c index df66450fb443..d4911a6e8331 100644 --- a/fs/file_table.c +++ b/fs/file_table.c @@ -264,7 +264,9 @@ static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput); void fput(struct file *file) { - if (atomic_long_dec_and_test(&file->f_count)) { + long cnt = atomic_long_dec_return(&file->f_count); + WARN_ON(cnt < 0); + if (cnt == 0) { struct task_struct *task = current; if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) { -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/1650062 Title: Ubuntu16.04.01VM:Docker-Powervm aufs bad file panic while running tests in a docker container To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1650062/+subscriptions -- ubuntu-bugs mailing list ubuntu-bugs@lists.ubuntu.com https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs